Homogeneous nucleation in metal vapors. 2. Dependence of the heat

Vitali V. Lissianski, Peter M. Maly, and Vladimir M. Zamansky , William C. Gardiner. Industrial & Engineering Chemistry Research 2001 40 (15), 3287-32...
0 downloads 0 Views 783KB Size
994

H. J. Freund and S. H. Bauer

Homogeneous Nucleation in Metal Vapors. 2. Dependence of the Heat of Condensation on Cluster Size H. J. Freund and S. H. Bauer' Department of Chemistry, Cornel1 Universlty, Ithaca, New York 14853 (Received July 19, 1976; Revised Manuscript Received March 7, 1977)

By shock heating moderately volatile metal bearing compounds [for example, Fe(CO)5]highly diluted in argon, supersaturated metal vapors, S = 50-3000, at controlled densities and temperatures can be generated. The subsequenthomogeneous condensation was followed by recording the turbidity of the gas as a function of distance behind the shock front. Determination of density gradients behind the shock front, using the laser-schlieren technique, provided measures of the rates and enthalpies of the exothermic reactions which occurred; in this case, the condensation of metal droplets. A model was developed which accounted for the observed density gradient profiles, and led to parameters which specify the particle size distribution as a function of time, and the average size (40-2000 atoms) to which the particles grow for any set of experimental conditions. These experiments also provide an estimate for the "binding energy" for the reaction nFe Fen, AE,/n = AEJ1 - n4,25),where A& is the bulk energy of sublimation. This empirical representation is surprisingly well fitted by several theoretical calculations of binding energies of small clusters (2 I n I 2000). +

Introduction In the first paper of this series' we described a shocktube technique for preparing controlled levels of supersaturated iron vapor, and procedures for determining the dependence of the critical supersaturation on the temperature. The catastrophic onset of condensation was ascertained by recording either a rapid rise in the turbidity, or the appearance of black body emission from the glowing clusters. These investigations have now been extended, and more refined diagnostics were developed. The starting point of the present report was the observation by Kung and Bauer' that the nascent clusters had a black body temperature which was always higher than that of the carrier gas (50-150 K); clearly there was a lag in the transfer of the heat of condensation to the ambient medium. The goal of our current study is the development of sufficient statistical-thermodynamic data so that a purely kinetic molecular model could be developed without recourse to bulk-type properties; for example, we consider it unacceptable to discuss the free-energy increment for cluster formation in terms of a surface free energy for very small aggregates (range 5 6 n 6 501, whereby one eventually substitutes bulk liquid magnitudes for the parameters of state. Here we take the first step, measurement of the enthalpy contribution to the free energy of cluster formation; in paper 5 we present estimates of the entropy term, and introduce a consistent kinetic model. During the first decade of this century J. J. Thomson discussed the condensation from a supersaturated vapor in terms of a metastable equilibrium between two phases. The nascent clusters were treated as tiny drops of liquid, to which the Kelvin equation was applied. The metaequilibrium vapor pressure (p,) for a uniform assembly of such droplets (radius r,), relative to that of the bulk liquid ( p , ) is

LT, and p, are the surface free energy and the monomer density, respectively, in a cluster size n. After six decades of evolution of theories for homogeneous condensation3this aspect of the model has not been fully relinquished, in spite of the fact that relation 1 connects obviously nonmeasurable quantities. One has but to inspect the interesting

The Journal of Physical Chemistry, Vol. 81, No. 10, 1977

density profiles of relatively large clusters derived by Lee et alS4who performed a Monte-Carlo energy minimization calculation. Inspection of their Figure 9 (radial density curves for 13,43, and 87 atom clusters) shows that there is no definable "surface" layer, nor can p, be defined, since p(r) decreases gradually to zero, at a radius determined by an arbitrarily specified confining volume, The obvious alternative to the application of bulk liquid concepts to clusters in the range 5 5 n 6 50 is to treat them 'as large molecules. We undertook to measure their heats of condensation from monomers. To our knowledge this is the first such reported measurement, and while at this stage our data are crude, they are of sufficient accuracy to affiim several novel conclusions. Indirect estimates have been made on the basis of mass spectrometric analyses of clusters from effusion cells or from divergent nozzle flows, but values for n > 10 are not generally available. Gingerich and co-workers reported extensively on mass spectrometric measurements of metallic clusters which effuse from Knudsen cells.5 They measured6 relative populations of tin clusters (1 < n < 8) over the temperature range 1379-1651 K, but to deduce AHo, it was necessary to introduce third law estimates for the entropies (ASo,). Configurational contributions were neglected. For lead (1 < n < 5) both second and third law values were ~ b t a i n e d . ~ However, the accuracy of data on the relative concentrations of clusters from freely expanding jets, based on mass s ectrometric detection, has been seriously questioned. There have been many theoretical studies of the dependence of energy content on cluster size. Among the recent reports, Andersong using a low spin MO approximation estimated the binding energies for 2 13 atom clusters of Ti, Cr, Fe, and Ni. Baetzold" undertook extended Huckel and CNDO MO calculations; he obtained cluster binding energies for Ag and Ni. He considered the effect of dimensionality on the binding energy, and found that for small numbers of metals atoms, a linear configuration is more stable than a three-dimensional structure. Fripiat et al.'* computed binding energies in Li crystals (at a fixed geometry) using the SCF X a scattered wave method. Another technique was introduced by Gelb et al.I2 (the diatomics-in-molecule approach) in their study of small clusters of Na atoms; Dykstra and S~haefer'~ applied the theory of self-consistent electron pairs to Bel. Burton'4

t

-

995

Homogeneous Nucleation in Metal Vapors

4 Laser

REDUCED SUBLIMATION ENERGIES I O

nA = A n

Shock tube

/-

a

Jo" 0

04

Fe (Anderson)

0 Ar IBurfonI

A

Aq (Baetzoldl

0

No IGeIbl

I L I IFripioI)

AA

0

A

I

I

I

IO

IO2

1

lo3

io4

-8-J Laser Schlieren

00, Figure 1. Theoretical (5 sets) and experimental values (1 set) for the reduced energy of condensation (per atom), as a function of cluster size. In the least-squares fitting of the points to a function of the form A€,,ln = A€,(1 - an-b). Baetzoid's values for Ag (three-dimensional calculations) were not included, but all other points were given equal weight. A correlation coefficient of 0.91 was obtained for best fit: a = 0.96 and b = 0.235. When b 0.250, the correlation coefficient declined inappreciably (to 0.90),for a = 1.0l5.Other computed points (including water) fall within a band of width fO.l about the empirical curve; for example, J. C. Owicki, et al., J. Phys. Chem.. 79, 1794 (1975).

determined the binding energies for small spherical clusters of argon using a normal mode analysis with energy minimization, based on a Lennard-Jones pair potential. These theoretical data are plotted on a reduced scale in Figure 1. [hE,/n] is the condensation energy per atom; AE, is the corresponding energy released upon condensation to a macroscopic cluster. In view of the density profiles within small clusters4 it is not surprising that the binding energy per atom for small n is considerably less than that for the bulk.I5 The question to be answered is the general shape of the function which ties the n = 2 to the n = lo4 points. Experimental Section We have demonstrated' that there was a measurable time lag for transfer of the heat of condensation from the nascent clusters to the carrier gas. It follows that the heating rate is determined by the following: (i) the nmer growth rate, averaged over the instantaneous size distribution; (ii) the differential heat of condensation for monomer addition; and (iii) the heat transfer coefficient from the nascent clusters to the ambient argon. The thermal balance downstream from a shock front can be measured by the very sensitive "laser-schlieren" technique developed by Kiefer and Lutz" for their study of the vibrational relaxation time of H2 and D2.Since the shock heated gas remains essentially at constant pressure, it follows that 1 d_pa 1dT _ pdx T dx Now, the index of refraction (m)is a linear function ofthe mass density (p): m 1 r(p/po),where r is the DaleGladstone constant, which is additive for a mixture of gases: r = Ci&ri;& is the mole fraction, and po is the reference density. Furthermore, the gradient in the index of refraction can be sensitively measured by the deflection of a narrow laser beam, using a long lever arm. Hence, the laser beam deflection is directly related to the rate of heat injection or removal from the carrier gas. The light is bent into the region of higher density. The density gradient is given in terms of the beam displacement 6 (see Figure 2):

+

aplax = GIwrL (2) For these experiments shock conditions were chosen so that the Fe(C0)5decomposed essentially within the shock

1 System

Figure 2. Schematic for the laser-schlieren system.

front. The estimated lifetime of the carbonyl a t 1600 K is less than 10-l' s; its initial concentration was 0.5-1.2% of the argon. Hence the average molecular weight of the test gas changed inappreciably upon dissociation of the Fe(CO)6or condensation of the Fe. Upon differentiation of the (ideal) equation of state, subject t_o the shock conservation equations, one can derive (R is the gas constant per unit mass):

(3) The subscripts 1 and 2 refer to the unshockd and shocked conditions, respectively. The magnitude of R is unchanged since the increment in molecular weight is small; u1is the shock speed. The local density behind the shock front is related to the corresponding temperature, and the initial conditions:

T2

(4) At any t after shock arrival, ap/ax is known; then following an interative procedure one can derive corresponding values for aT/ax. In turn, the temperature gradient is determined by the change in the specific enthalpy of the €35

(5) where cp is the heat capacity of the gas (here also we neglected the small effect of the Fe), and u2 is the shocked gas velocity. Integration with respect to time allows one to estimate the total change in the enthalpy (or temperature) due to condensation of the Fe atoms. An estimate of the sensitivity of our laser-schlieren system, expressed in terms of a general mechanism wherein several reactions occur c~ncurrently,~' is ( + / a x ) a 2 [ ~ ~ ( -f )62pq a q o j

2 0.02 kcal crn-js-' (6) where and GI?) are the forward and reverse rates for the j t h reaction, in mol s-l, and AHjo is the corresponding net enthalpy change per mole. Experimental Details and Data Figure 3 is a schematic of the overall experimental arrangement. Photomultiplier 2 records the schlieren The Journal of Physical Chemistry, Vol. 81, No. 10, 1977

996

H. J. Freund and S. H. Bauer

TABLE I: Typical Experimental Parameters

T,= 298 K

T,= 1600-1800 K (2.3-4.6) X g cm-3 p , = (1-2) X g cm-) P,= (10-20 ) TOIT P, = (0.3-0.7) atm (M) = 39.7 g mol-’ u, = (1.37-1.46) mm ps-’ Post dissociation of Fe(CO),: mol % Fe = 1.15 ap/ax = (10-7-10-5)g cm-4 dT,/dt= (0.1-10) K ps-l dh,/dt= (1-8) x l o 3 erg ~ r n ~- s~- l p1=

\ \ (To Channel il

\

\

Schlieren perfurbation

k!

HV

244cm

Supply

(To

Channel 2)

Figure 4. Dual beam scope traces: (a) schlieren slgnal, 0.05 V/dMSh, (b) turbidity; 0.02 V/dMsion. The zero level for this beam was displaced above the screen to permit recording of the small intensity loss. For this run, T2 = 1648 K; p 2 = 1.81 X loi6 monomers ~ m - ~ .

I

n Fe = Fen

Q=

Flgure 3. Schematic of overall experimental arrangement: F, and F2 are a neutral density filter and narrow band filter, respectively; P2 to pumps, sample reservoir, and pressure gauges: M2 is mounted on a rotatable shaft for intensity level calibration. The dlstance from the center of the shock tube to the rotating mirror is 555 cm.

Av. condensation enthalpy depends on MEAN CLUSTER SIZE

(NFe)

--I

h

signal while photomultiplier 1 measures the attenuation of the beam by the turbidity of the shocked sample due to absorption and scattering by the growing clusters. This channel provides the necessary correction ratio for the schlieren signal, and a measure of the amount of material which had condensed at any time. The shock tube is of stainless steel, of rectangular cross section, 6.35 cm by 4.45 cm. Shock waves were initiated by pressure bursting prescribed 16-mil aluminum diaphragms; the driver gas was He or He/N2mixtures. Before each run, the driven section was pumped to below Torr. The combined leak-outgassing rate was 5 X Torr/min. Approximately 1 min was taken to fill the tube with the sample and to burst the diaphragm. Sample loading pressures ranged between 10 and 20 Torr. The Fe(C0)5was double distilled, and each time we retained only the middle fraction; ultra high purity argon was used as the diluent. The experiments were made with 0.5 or 1.22 mol % of the carbonyl. Shock speeds were measured using two platinum film heat transfer gauges (no. 3 and 4 in Figure 3). Shock attenuation was found to be about 1% /m, but ideal shock equations were used to calculate the temperature. The light source was a Spectra-Physics Model 123 He-Ne laser, equipped with a model 235A exciter; its output power was specified at 7 mW. The beam width at the center of the shock tube was 0.9 mm a t the e-2 points. An EM1 9558 phototube was used in channel 1and EM1 9658 in channel 2. Both phototubes have a l / e response time of less than 200 ns. The calibration procedure via the rotating mirror has been described in a previous publication.18 Initial and typical shock heated conditions are listed in Table I. Figure 4 shows an oscilloscope record; (a) measures the density gradient which must be corrected for the beam attenuation using the lower trace (b). The latter is a measure of the change with time of the fraction of material condensed. Reduced graphs are illustrated in Figure 5 for The Journal of Physical Chemistry, Vol. 81, No. IO, 1977

. 0

3.

t (psec ,particle time)

Flgure 5. Reduced curves for the rate of temperature rise and the integrated increment in the temperature of the carrier gas.

run B-5. Via iterative calculations with eq 3-5, the total heat injected into the Ar per unit volume, Q, can be determined from the corrected density gradient trace. This quantity divided by the initial number density of iron atoms gives the heat released per Fe atom ( q ) due to the exothermic condensation step. These results are summarized in Table 11. Our initial assumption was that a narrow distribution of clusters, average size (n),is present during condensation. The reaction (n)Feg -+ Feb)

(7 )

expresses the “phase” transformation which occurs. The mean enthalpy increment, per atom, for this reaction (or, the mean atomization energy per atom) may be identified with the experimentally determined quantity, q. The obvious check is satisfactory; the q values are in the range (4-6) X erg/atom, somewhat less than but close to the heat of condensation (per atom) to the bulk phase. For a distribution of finite width we must estimate an average

QQ7

Homogeneous Nucleation In Metal Vapors

TABLE 11: Experimental Parameters and Deduced Magnitudes of Heat Release Run no, B-8 B-33 B-5 B-11 B-24 A-24 A-27 A-21 A-18

P I(tota1 ), Torr

ulQ

T2 ( t =0), K

10.4 10.3 10.2 10.7 10.3 20.4 20.4 20.8 20.2

1.377 1.431 1.405 1.374 1.416 1.449 1.458 1.425 1.416

1617 1747 1684 1609 1710 1791 1814 1732 1710

Shock speed, in mm ps-'.

T,= 298

NFe

( t 7 o),

cm1.80 X 10l6 1.82 X 10l6 1.81 X 10l6 1.89 1.82 3.62 3.60 3.70 3.59

X

X

1Ol6

loL6

X 10l6 X 10" X 10l6

x 10l6 K,C, = 0.131 cd-' g K-',

size ( n ) , for the different supersaturation ratios. Then we could determine the effect of cluster size on the enthalpy of the condensation reaction. To obtain ( n ) , a specific model has to be inserted.

Energy Transfer Model The recorded beam deflection measures the rate of change of the enthalpy content of the gas sample probed by the beam; i.e., the rate of injection of energy into the carrier gas behind the incident shock wave, dhg/dt. We sought the simplest model which reproduces the shape of these (dh,/dt) traces. We shall point to the approximations which were made in the model and hope either to remove them or to justify them in depth at the next stage of its evolution. In the present form, four parameters were introduced which, upon evaluation to optimize the fit, relate the binding energy of small clusters to their size, and specify the change of the distribution function of sizes with time. The rate of change of the energy content of the clusters, as they grow and concurrently cool, is the sum of two principal terms. The first is the rate of increase of its energy content due to monomer addition, and the second term is the loss of energy due to collisions with the cooler ambient gas. We neglected loss of energy due to radiation. The following definitions were used: n number of atoms per cluster; n will be treated as a continuous variable E,(n,t) mean energy content at time t of an nmer Nl(t) monomer density at time t Ng carrier gas density E, average speed of carrier gas molecules, averaged over the temperature range for the run El average speed of the monomer A(n) surface area of nmer (assumed spherical) a(n) net sticking coefficient r(n) energy increment for the addition of a monomer: + Fe Fen T, ambient gas temperature e(n) fraction of excess energy transferred from nmer to the carrier gas, per collision with a gas molecule Then the rate of change of energy content of the clusters is given by dE,/dt= ' / 4 {N1(t)c1A(n)cu(n)[3/2hT, + r(n)]

-

- N g C p A ( n ) W ) [ E , ( n ,t ) - E,@, L)l)

(8) We did not include a term for cooling due to evaporation; Le., the loss of a monomer either spontaneously or induced by a sufficiently energetic collision with an Ar atom. The latter process is of low probability, but the former may be significant. To take this into consideration requires a considerably more complex model than is used in the present analysis. l9 However, in the accompanying paper (part 5) we demonstrated that during cluster growth for Fe evaporation is negligible for n >, 15. In addition, we

Sb

AT, K

2250 310 773 2640 536 344 307 796 1050

130 143 190 135

181 216 238 115 184

Q, erg cmF3 7.11 X l o 4 7.95X l o 4 9.88 x io4 7.64 X lo4 9.44 x io4 2.22 x 105 2.38X lo5 1.29X l o 5 2.01 X l o 5

q , erg atom-' 3.95 x 4.36 X lo-'' 5.46 X 10"' 4.04 X lo-'' 5.2 X lo-'* 6.1 X lo-'' 6.6 X 3.47 x 5.60 X lo-''

lo-'*

s is the supersaturation ratio at t = 0. s = PFe/PFeeq*

assumed: (i) the second term of the right member of (8) is the net heat transfer due to collisions between clusters and the ambient gas; (ii) the gas temperature, Tg,and the number density, N ,are constant with time. Although the temperature and t8e density obviously do change due to condensation, the fractional change in this dilute system is small; thus in (8), average values of the temperature and density are implied, (iii) in the following analysis we neglect (with justification) the translational energy, ( 3 / 2 ) k T g , compared to I'(n);finally (iv) in formulating the cooling term, we assumed that d clusters of size n (at any instant) have the same (mean) energy content E&, t). The introduction of another parameterized distribution function at this stage would hopelessly complicate the analysis. Tests of a hybrid model in which we introduced a transition layer between the hot cluster and the ambient gas showed that, to a very good approximation, the gas temperature in this layer was equal to the temperature of the surroundings. Equation 8 can be simplified by writing (for t > 0):

a) E

nN c --%9[3uc/(4n)]'/3 4

(9)

where u, is the volume per atom in the condensed phase, at the average temperature. Thus dY(n, t)/dt = (p(t)n2l3r ( n ) a ( n )

-a

e ( n ) ~ ( nt ),

) t ~ ~ / ~

(10)

This equation has an analytic solution if q(t) is specified. We may deduce dt)from N,(t) by empirically fitting the observed turbidity curve to a polynomial of the form ( A + Bt + Ct2 + Dt3),between t = 0 and the time at which the trace levels off (end of condensation). Since the attenuation [(I: - 11)a ~ ( t measures )] the cluster density:" one may thus estimate the concentration of the remaining monomer." The constant of integration for (10) is evaluated by setting Y(n,0 ) = 0, since at t = 0 no nmer had been generated. The rate of energy transfer to the ambient gas is the product of the second term in eq 8 and the cluster distribution function, integrated over n: r:t

) = a).f;dn N(n, t)n2I3O(n)Y(n, t)

where we have identified the specific enthalpy with the energy transferred. N(n, t) is defined as the number density of nmers, and we have assumed it has the general formz2

N(n, t ) = f n exp(-gn') The Journal of PhplC8l Chemlstry, Vol. 8 1 , No.

(12) IO, 1977

998

H. J. Freund and S. H. Bauer

The conservation condition requires

Jy N ( n , t ) n dn = N(1, O ) x ( t )

TABLE 111: Values of tn), Estimated by Fitting the Model to Densits Gradients

(13)

where x ( t ) is the fraction of starting material which entered the condensed phase. The choice of this functional form is not overly restrictive in so far as these experiments are concerned. Indeed, a delta function was also tested, and while the overall fit for the experimental runs was not quite as good as with (12) it did indicate that the model is not particularly sensitive to the precise form of the distribution function. The normalization condition leads to f = (41 ~ ' / ' ) g ~ / ~ NO)x(t), ( l , when the lower limit is set at zero (approximately). The median size ( n )can be determined graphically from

Jp'nN(n, t ) dn = JG)nN(n,t ) dn

(14)

The maximum of [nN(n,t ) ] is at g'/2. To a very good approximation n at N,, is equal to the median size. Since g is a function of time, the average size is then specified by ( n ( t ) )= [ g ( t ) ] - ' / 2 ;thus, g ( t ) characterizes the time dependence of the condensation process. For this function we assumed a simple form g ( t ) = [l

+ xP(t)